ivshmem: Fix unplug of device "ivshmem-plain"
[qemu/armbru.git] / migration / migration.c
blobd6ae879dc8d7c625c926c26b86927ed835a8bfa9
1 /*
2 * QEMU live migration
4 * Copyright IBM, Corp. 2008
6 * Authors:
7 * Anthony Liguori <aliguori@us.ibm.com>
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
12 * Contributions after 2012-01-13 are licensed under the terms of the
13 * GNU GPL, version 2 or (at your option) any later version.
16 #include "qemu/osdep.h"
17 #include "qemu/cutils.h"
18 #include "qemu/error-report.h"
19 #include "migration/blocker.h"
20 #include "exec.h"
21 #include "fd.h"
22 #include "socket.h"
23 #include "rdma.h"
24 #include "ram.h"
25 #include "migration/global_state.h"
26 #include "migration/misc.h"
27 #include "migration.h"
28 #include "savevm.h"
29 #include "qemu-file-channel.h"
30 #include "qemu-file.h"
31 #include "migration/vmstate.h"
32 #include "block/block.h"
33 #include "qapi/error.h"
34 #include "qapi/qapi-commands-migration.h"
35 #include "qapi/qapi-events-migration.h"
36 #include "qapi/qmp/qerror.h"
37 #include "qapi/qmp/qnull.h"
38 #include "qemu/rcu.h"
39 #include "block.h"
40 #include "postcopy-ram.h"
41 #include "qemu/thread.h"
42 #include "trace.h"
43 #include "exec/target_page.h"
44 #include "io/channel-buffer.h"
45 #include "migration/colo.h"
46 #include "hw/boards.h"
47 #include "monitor/monitor.h"
49 #define MAX_THROTTLE (32 << 20) /* Migration transfer speed throttling */
51 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
52 * data. */
53 #define BUFFER_DELAY 100
54 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
56 /* Time in milliseconds we are allowed to stop the source,
57 * for sending the last part */
58 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
60 /* Maximum migrate downtime set to 2000 seconds */
61 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000
62 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
64 /* Default compression thread count */
65 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
66 /* Default decompression thread count, usually decompression is at
67 * least 4 times as fast as compression.*/
68 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
69 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
70 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
71 /* Define default autoconverge cpu throttle migration parameters */
72 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
73 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
74 #define DEFAULT_MIGRATE_MAX_CPU_THROTTLE 99
76 /* Migration XBZRLE default cache size */
77 #define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024)
79 /* The delay time (in ms) between two COLO checkpoints
80 * Note: Please change this default value to 10000 when we support hybrid mode.
82 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY 200
83 #define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2
84 #define DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT 16
86 /* Background transfer rate for postcopy, 0 means unlimited, note
87 * that page requests can still exceed this limit.
89 #define DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH 0
91 static NotifierList migration_state_notifiers =
92 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
94 static bool deferred_incoming;
96 /* Messages sent on the return path from destination to source */
97 enum mig_rp_message_type {
98 MIG_RP_MSG_INVALID = 0, /* Must be 0 */
99 MIG_RP_MSG_SHUT, /* sibling will not send any more RP messages */
100 MIG_RP_MSG_PONG, /* Response to a PING; data (seq: be32 ) */
102 MIG_RP_MSG_REQ_PAGES_ID, /* data (start: be64, len: be32, id: string) */
103 MIG_RP_MSG_REQ_PAGES, /* data (start: be64, len: be32) */
104 MIG_RP_MSG_RECV_BITMAP, /* send recved_bitmap back to source */
105 MIG_RP_MSG_RESUME_ACK, /* tell source that we are ready to resume */
107 MIG_RP_MSG_MAX
110 /* When we add fault tolerance, we could have several
111 migrations at once. For now we don't need to add
112 dynamic creation of migration */
114 static MigrationState *current_migration;
115 static MigrationIncomingState *current_incoming;
117 static bool migration_object_check(MigrationState *ms, Error **errp);
118 static int migration_maybe_pause(MigrationState *s,
119 int *current_active_state,
120 int new_state);
122 void migration_object_init(void)
124 MachineState *ms = MACHINE(qdev_get_machine());
125 Error *err = NULL;
127 /* This can only be called once. */
128 assert(!current_migration);
129 current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION));
132 * Init the migrate incoming object as well no matter whether
133 * we'll use it or not.
135 assert(!current_incoming);
136 current_incoming = g_new0(MigrationIncomingState, 1);
137 current_incoming->state = MIGRATION_STATUS_NONE;
138 current_incoming->postcopy_remote_fds =
139 g_array_new(FALSE, TRUE, sizeof(struct PostCopyFD));
140 qemu_mutex_init(&current_incoming->rp_mutex);
141 qemu_event_init(&current_incoming->main_thread_load_event, false);
142 qemu_sem_init(&current_incoming->postcopy_pause_sem_dst, 0);
143 qemu_sem_init(&current_incoming->postcopy_pause_sem_fault, 0);
145 init_dirty_bitmap_incoming_migration();
147 if (!migration_object_check(current_migration, &err)) {
148 error_report_err(err);
149 exit(1);
153 * We cannot really do this in migration_instance_init() since at
154 * that time global properties are not yet applied, then this
155 * value will be definitely replaced by something else.
157 if (ms->enforce_config_section) {
158 current_migration->send_configuration = true;
162 void migration_object_finalize(void)
164 object_unref(OBJECT(current_migration));
167 /* For outgoing */
168 MigrationState *migrate_get_current(void)
170 /* This can only be called after the object created. */
171 assert(current_migration);
172 return current_migration;
175 MigrationIncomingState *migration_incoming_get_current(void)
177 assert(current_incoming);
178 return current_incoming;
181 void migration_incoming_state_destroy(void)
183 struct MigrationIncomingState *mis = migration_incoming_get_current();
185 if (mis->to_src_file) {
186 /* Tell source that we are done */
187 migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0);
188 qemu_fclose(mis->to_src_file);
189 mis->to_src_file = NULL;
192 if (mis->from_src_file) {
193 qemu_fclose(mis->from_src_file);
194 mis->from_src_file = NULL;
196 if (mis->postcopy_remote_fds) {
197 g_array_free(mis->postcopy_remote_fds, TRUE);
198 mis->postcopy_remote_fds = NULL;
201 qemu_event_reset(&mis->main_thread_load_event);
204 static void migrate_generate_event(int new_state)
206 if (migrate_use_events()) {
207 qapi_event_send_migration(new_state);
211 static bool migrate_late_block_activate(void)
213 MigrationState *s;
215 s = migrate_get_current();
217 return s->enabled_capabilities[
218 MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE];
222 * Called on -incoming with a defer: uri.
223 * The migration can be started later after any parameters have been
224 * changed.
226 static void deferred_incoming_migration(Error **errp)
228 if (deferred_incoming) {
229 error_setg(errp, "Incoming migration already deferred");
231 deferred_incoming = true;
235 * Send a message on the return channel back to the source
236 * of the migration.
238 static int migrate_send_rp_message(MigrationIncomingState *mis,
239 enum mig_rp_message_type message_type,
240 uint16_t len, void *data)
242 int ret = 0;
244 trace_migrate_send_rp_message((int)message_type, len);
245 qemu_mutex_lock(&mis->rp_mutex);
248 * It's possible that the file handle got lost due to network
249 * failures.
251 if (!mis->to_src_file) {
252 ret = -EIO;
253 goto error;
256 qemu_put_be16(mis->to_src_file, (unsigned int)message_type);
257 qemu_put_be16(mis->to_src_file, len);
258 qemu_put_buffer(mis->to_src_file, data, len);
259 qemu_fflush(mis->to_src_file);
261 /* It's possible that qemu file got error during sending */
262 ret = qemu_file_get_error(mis->to_src_file);
264 error:
265 qemu_mutex_unlock(&mis->rp_mutex);
266 return ret;
269 /* Request a range of pages from the source VM at the given
270 * start address.
271 * rbname: Name of the RAMBlock to request the page in, if NULL it's the same
272 * as the last request (a name must have been given previously)
273 * Start: Address offset within the RB
274 * Len: Length in bytes required - must be a multiple of pagesize
276 int migrate_send_rp_req_pages(MigrationIncomingState *mis, const char *rbname,
277 ram_addr_t start, size_t len)
279 uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
280 size_t msglen = 12; /* start + len */
281 enum mig_rp_message_type msg_type;
283 *(uint64_t *)bufc = cpu_to_be64((uint64_t)start);
284 *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len);
286 if (rbname) {
287 int rbname_len = strlen(rbname);
288 assert(rbname_len < 256);
290 bufc[msglen++] = rbname_len;
291 memcpy(bufc + msglen, rbname, rbname_len);
292 msglen += rbname_len;
293 msg_type = MIG_RP_MSG_REQ_PAGES_ID;
294 } else {
295 msg_type = MIG_RP_MSG_REQ_PAGES;
298 return migrate_send_rp_message(mis, msg_type, msglen, bufc);
301 void qemu_start_incoming_migration(const char *uri, Error **errp)
303 const char *p;
305 qapi_event_send_migration(MIGRATION_STATUS_SETUP);
306 if (!strcmp(uri, "defer")) {
307 deferred_incoming_migration(errp);
308 } else if (strstart(uri, "tcp:", &p)) {
309 tcp_start_incoming_migration(p, errp);
310 #ifdef CONFIG_RDMA
311 } else if (strstart(uri, "rdma:", &p)) {
312 rdma_start_incoming_migration(p, errp);
313 #endif
314 } else if (strstart(uri, "exec:", &p)) {
315 exec_start_incoming_migration(p, errp);
316 } else if (strstart(uri, "unix:", &p)) {
317 unix_start_incoming_migration(p, errp);
318 } else if (strstart(uri, "fd:", &p)) {
319 fd_start_incoming_migration(p, errp);
320 } else {
321 error_setg(errp, "unknown migration protocol: %s", uri);
325 static void process_incoming_migration_bh(void *opaque)
327 Error *local_err = NULL;
328 MigrationIncomingState *mis = opaque;
330 /* If capability late_block_activate is set:
331 * Only fire up the block code now if we're going to restart the
332 * VM, else 'cont' will do it.
333 * This causes file locking to happen; so we don't want it to happen
334 * unless we really are starting the VM.
336 if (!migrate_late_block_activate() ||
337 (autostart && (!global_state_received() ||
338 global_state_get_runstate() == RUN_STATE_RUNNING))) {
339 /* Make sure all file formats flush their mutable metadata.
340 * If we get an error here, just don't restart the VM yet. */
341 bdrv_invalidate_cache_all(&local_err);
342 if (local_err) {
343 error_report_err(local_err);
344 local_err = NULL;
345 autostart = false;
350 * This must happen after all error conditions are dealt with and
351 * we're sure the VM is going to be running on this host.
353 qemu_announce_self();
355 if (multifd_load_cleanup(&local_err) != 0) {
356 error_report_err(local_err);
357 autostart = false;
359 /* If global state section was not received or we are in running
360 state, we need to obey autostart. Any other state is set with
361 runstate_set. */
363 dirty_bitmap_mig_before_vm_start();
365 if (!global_state_received() ||
366 global_state_get_runstate() == RUN_STATE_RUNNING) {
367 if (autostart) {
368 vm_start();
369 } else {
370 runstate_set(RUN_STATE_PAUSED);
372 } else {
373 runstate_set(global_state_get_runstate());
376 * This must happen after any state changes since as soon as an external
377 * observer sees this event they might start to prod at the VM assuming
378 * it's ready to use.
380 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
381 MIGRATION_STATUS_COMPLETED);
382 qemu_bh_delete(mis->bh);
383 migration_incoming_state_destroy();
386 static void process_incoming_migration_co(void *opaque)
388 MigrationIncomingState *mis = migration_incoming_get_current();
389 PostcopyState ps;
390 int ret;
392 assert(mis->from_src_file);
393 mis->migration_incoming_co = qemu_coroutine_self();
394 mis->largest_page_size = qemu_ram_pagesize_largest();
395 postcopy_state_set(POSTCOPY_INCOMING_NONE);
396 migrate_set_state(&mis->state, MIGRATION_STATUS_NONE,
397 MIGRATION_STATUS_ACTIVE);
398 ret = qemu_loadvm_state(mis->from_src_file);
400 ps = postcopy_state_get();
401 trace_process_incoming_migration_co_end(ret, ps);
402 if (ps != POSTCOPY_INCOMING_NONE) {
403 if (ps == POSTCOPY_INCOMING_ADVISE) {
405 * Where a migration had postcopy enabled (and thus went to advise)
406 * but managed to complete within the precopy period, we can use
407 * the normal exit.
409 postcopy_ram_incoming_cleanup(mis);
410 } else if (ret >= 0) {
412 * Postcopy was started, cleanup should happen at the end of the
413 * postcopy thread.
415 trace_process_incoming_migration_co_postcopy_end_main();
416 return;
418 /* Else if something went wrong then just fall out of the normal exit */
421 /* we get COLO info, and know if we are in COLO mode */
422 if (!ret && migration_incoming_enable_colo()) {
423 qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming",
424 colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE);
425 mis->have_colo_incoming_thread = true;
426 qemu_coroutine_yield();
428 /* Wait checkpoint incoming thread exit before free resource */
429 qemu_thread_join(&mis->colo_incoming_thread);
432 if (ret < 0) {
433 Error *local_err = NULL;
435 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
436 MIGRATION_STATUS_FAILED);
437 error_report("load of migration failed: %s", strerror(-ret));
438 qemu_fclose(mis->from_src_file);
439 if (multifd_load_cleanup(&local_err) != 0) {
440 error_report_err(local_err);
442 exit(EXIT_FAILURE);
444 mis->bh = qemu_bh_new(process_incoming_migration_bh, mis);
445 qemu_bh_schedule(mis->bh);
446 mis->migration_incoming_co = NULL;
449 static void migration_incoming_setup(QEMUFile *f)
451 MigrationIncomingState *mis = migration_incoming_get_current();
453 if (multifd_load_setup() != 0) {
454 /* We haven't been able to create multifd threads
455 nothing better to do */
456 exit(EXIT_FAILURE);
459 if (!mis->from_src_file) {
460 mis->from_src_file = f;
462 qemu_file_set_blocking(f, false);
465 void migration_incoming_process(void)
467 Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, NULL);
468 qemu_coroutine_enter(co);
471 /* Returns true if recovered from a paused migration, otherwise false */
472 static bool postcopy_try_recover(QEMUFile *f)
474 MigrationIncomingState *mis = migration_incoming_get_current();
476 if (mis->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
477 /* Resumed from a paused postcopy migration */
479 mis->from_src_file = f;
480 /* Postcopy has standalone thread to do vm load */
481 qemu_file_set_blocking(f, true);
483 /* Re-configure the return path */
484 mis->to_src_file = qemu_file_get_return_path(f);
486 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_PAUSED,
487 MIGRATION_STATUS_POSTCOPY_RECOVER);
490 * Here, we only wake up the main loading thread (while the
491 * fault thread will still be waiting), so that we can receive
492 * commands from source now, and answer it if needed. The
493 * fault thread will be woken up afterwards until we are sure
494 * that source is ready to reply to page requests.
496 qemu_sem_post(&mis->postcopy_pause_sem_dst);
497 return true;
500 return false;
503 void migration_fd_process_incoming(QEMUFile *f)
505 if (postcopy_try_recover(f)) {
506 return;
509 migration_incoming_setup(f);
510 migration_incoming_process();
513 void migration_ioc_process_incoming(QIOChannel *ioc)
515 MigrationIncomingState *mis = migration_incoming_get_current();
516 bool start_migration;
518 if (!mis->from_src_file) {
519 /* The first connection (multifd may have multiple) */
520 QEMUFile *f = qemu_fopen_channel_input(ioc);
522 /* If it's a recovery, we're done */
523 if (postcopy_try_recover(f)) {
524 return;
527 migration_incoming_setup(f);
530 * Common migration only needs one channel, so we can start
531 * right now. Multifd needs more than one channel, we wait.
533 start_migration = !migrate_use_multifd();
534 } else {
535 /* Multiple connections */
536 assert(migrate_use_multifd());
537 start_migration = multifd_recv_new_channel(ioc);
540 if (start_migration) {
541 migration_incoming_process();
546 * @migration_has_all_channels: We have received all channels that we need
548 * Returns true when we have got connections to all the channels that
549 * we need for migration.
551 bool migration_has_all_channels(void)
553 MigrationIncomingState *mis = migration_incoming_get_current();
554 bool all_channels;
556 all_channels = multifd_recv_all_channels_created();
558 return all_channels && mis->from_src_file != NULL;
562 * Send a 'SHUT' message on the return channel with the given value
563 * to indicate that we've finished with the RP. Non-0 value indicates
564 * error.
566 void migrate_send_rp_shut(MigrationIncomingState *mis,
567 uint32_t value)
569 uint32_t buf;
571 buf = cpu_to_be32(value);
572 migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf);
576 * Send a 'PONG' message on the return channel with the given value
577 * (normally in response to a 'PING')
579 void migrate_send_rp_pong(MigrationIncomingState *mis,
580 uint32_t value)
582 uint32_t buf;
584 buf = cpu_to_be32(value);
585 migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf);
588 void migrate_send_rp_recv_bitmap(MigrationIncomingState *mis,
589 char *block_name)
591 char buf[512];
592 int len;
593 int64_t res;
596 * First, we send the header part. It contains only the len of
597 * idstr, and the idstr itself.
599 len = strlen(block_name);
600 buf[0] = len;
601 memcpy(buf + 1, block_name, len);
603 if (mis->state != MIGRATION_STATUS_POSTCOPY_RECOVER) {
604 error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
605 __func__);
606 return;
609 migrate_send_rp_message(mis, MIG_RP_MSG_RECV_BITMAP, len + 1, buf);
612 * Next, we dump the received bitmap to the stream.
614 * TODO: currently we are safe since we are the only one that is
615 * using the to_src_file handle (fault thread is still paused),
616 * and it's ok even not taking the mutex. However the best way is
617 * to take the lock before sending the message header, and release
618 * the lock after sending the bitmap.
620 qemu_mutex_lock(&mis->rp_mutex);
621 res = ramblock_recv_bitmap_send(mis->to_src_file, block_name);
622 qemu_mutex_unlock(&mis->rp_mutex);
624 trace_migrate_send_rp_recv_bitmap(block_name, res);
627 void migrate_send_rp_resume_ack(MigrationIncomingState *mis, uint32_t value)
629 uint32_t buf;
631 buf = cpu_to_be32(value);
632 migrate_send_rp_message(mis, MIG_RP_MSG_RESUME_ACK, sizeof(buf), &buf);
635 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
637 MigrationCapabilityStatusList *head = NULL;
638 MigrationCapabilityStatusList *caps;
639 MigrationState *s = migrate_get_current();
640 int i;
642 caps = NULL; /* silence compiler warning */
643 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
644 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
645 if (i == MIGRATION_CAPABILITY_BLOCK) {
646 continue;
648 #endif
649 if (head == NULL) {
650 head = g_malloc0(sizeof(*caps));
651 caps = head;
652 } else {
653 caps->next = g_malloc0(sizeof(*caps));
654 caps = caps->next;
656 caps->value =
657 g_malloc(sizeof(*caps->value));
658 caps->value->capability = i;
659 caps->value->state = s->enabled_capabilities[i];
662 return head;
665 MigrationParameters *qmp_query_migrate_parameters(Error **errp)
667 MigrationParameters *params;
668 MigrationState *s = migrate_get_current();
670 /* TODO use QAPI_CLONE() instead of duplicating it inline */
671 params = g_malloc0(sizeof(*params));
672 params->has_compress_level = true;
673 params->compress_level = s->parameters.compress_level;
674 params->has_compress_threads = true;
675 params->compress_threads = s->parameters.compress_threads;
676 params->has_compress_wait_thread = true;
677 params->compress_wait_thread = s->parameters.compress_wait_thread;
678 params->has_decompress_threads = true;
679 params->decompress_threads = s->parameters.decompress_threads;
680 params->has_cpu_throttle_initial = true;
681 params->cpu_throttle_initial = s->parameters.cpu_throttle_initial;
682 params->has_cpu_throttle_increment = true;
683 params->cpu_throttle_increment = s->parameters.cpu_throttle_increment;
684 params->has_tls_creds = true;
685 params->tls_creds = g_strdup(s->parameters.tls_creds);
686 params->has_tls_hostname = true;
687 params->tls_hostname = g_strdup(s->parameters.tls_hostname);
688 params->has_max_bandwidth = true;
689 params->max_bandwidth = s->parameters.max_bandwidth;
690 params->has_downtime_limit = true;
691 params->downtime_limit = s->parameters.downtime_limit;
692 params->has_x_checkpoint_delay = true;
693 params->x_checkpoint_delay = s->parameters.x_checkpoint_delay;
694 params->has_block_incremental = true;
695 params->block_incremental = s->parameters.block_incremental;
696 params->has_x_multifd_channels = true;
697 params->x_multifd_channels = s->parameters.x_multifd_channels;
698 params->has_x_multifd_page_count = true;
699 params->x_multifd_page_count = s->parameters.x_multifd_page_count;
700 params->has_xbzrle_cache_size = true;
701 params->xbzrle_cache_size = s->parameters.xbzrle_cache_size;
702 params->has_max_postcopy_bandwidth = true;
703 params->max_postcopy_bandwidth = s->parameters.max_postcopy_bandwidth;
704 params->has_max_cpu_throttle = true;
705 params->max_cpu_throttle = s->parameters.max_cpu_throttle;
707 return params;
711 * Return true if we're already in the middle of a migration
712 * (i.e. any of the active or setup states)
714 static bool migration_is_setup_or_active(int state)
716 switch (state) {
717 case MIGRATION_STATUS_ACTIVE:
718 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
719 case MIGRATION_STATUS_POSTCOPY_PAUSED:
720 case MIGRATION_STATUS_POSTCOPY_RECOVER:
721 case MIGRATION_STATUS_SETUP:
722 case MIGRATION_STATUS_PRE_SWITCHOVER:
723 case MIGRATION_STATUS_DEVICE:
724 return true;
726 default:
727 return false;
732 static void populate_ram_info(MigrationInfo *info, MigrationState *s)
734 info->has_ram = true;
735 info->ram = g_malloc0(sizeof(*info->ram));
736 info->ram->transferred = ram_counters.transferred;
737 info->ram->total = ram_bytes_total();
738 info->ram->duplicate = ram_counters.duplicate;
739 /* legacy value. It is not used anymore */
740 info->ram->skipped = 0;
741 info->ram->normal = ram_counters.normal;
742 info->ram->normal_bytes = ram_counters.normal *
743 qemu_target_page_size();
744 info->ram->mbps = s->mbps;
745 info->ram->dirty_sync_count = ram_counters.dirty_sync_count;
746 info->ram->postcopy_requests = ram_counters.postcopy_requests;
747 info->ram->page_size = qemu_target_page_size();
748 info->ram->multifd_bytes = ram_counters.multifd_bytes;
750 if (migrate_use_xbzrle()) {
751 info->has_xbzrle_cache = true;
752 info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
753 info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
754 info->xbzrle_cache->bytes = xbzrle_counters.bytes;
755 info->xbzrle_cache->pages = xbzrle_counters.pages;
756 info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss;
757 info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate;
758 info->xbzrle_cache->overflow = xbzrle_counters.overflow;
761 if (migrate_use_compression()) {
762 info->has_compression = true;
763 info->compression = g_malloc0(sizeof(*info->compression));
764 info->compression->pages = compression_counters.pages;
765 info->compression->busy = compression_counters.busy;
766 info->compression->busy_rate = compression_counters.busy_rate;
767 info->compression->compressed_size =
768 compression_counters.compressed_size;
769 info->compression->compression_rate =
770 compression_counters.compression_rate;
773 if (cpu_throttle_active()) {
774 info->has_cpu_throttle_percentage = true;
775 info->cpu_throttle_percentage = cpu_throttle_get_percentage();
778 if (s->state != MIGRATION_STATUS_COMPLETED) {
779 info->ram->remaining = ram_bytes_remaining();
780 info->ram->dirty_pages_rate = ram_counters.dirty_pages_rate;
784 static void populate_disk_info(MigrationInfo *info)
786 if (blk_mig_active()) {
787 info->has_disk = true;
788 info->disk = g_malloc0(sizeof(*info->disk));
789 info->disk->transferred = blk_mig_bytes_transferred();
790 info->disk->remaining = blk_mig_bytes_remaining();
791 info->disk->total = blk_mig_bytes_total();
795 static void fill_source_migration_info(MigrationInfo *info)
797 MigrationState *s = migrate_get_current();
799 switch (s->state) {
800 case MIGRATION_STATUS_NONE:
801 /* no migration has happened ever */
802 /* do not overwrite destination migration status */
803 return;
804 break;
805 case MIGRATION_STATUS_SETUP:
806 info->has_status = true;
807 info->has_total_time = false;
808 break;
809 case MIGRATION_STATUS_ACTIVE:
810 case MIGRATION_STATUS_CANCELLING:
811 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
812 case MIGRATION_STATUS_PRE_SWITCHOVER:
813 case MIGRATION_STATUS_DEVICE:
814 case MIGRATION_STATUS_POSTCOPY_PAUSED:
815 case MIGRATION_STATUS_POSTCOPY_RECOVER:
816 /* TODO add some postcopy stats */
817 info->has_status = true;
818 info->has_total_time = true;
819 info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
820 - s->start_time;
821 info->has_expected_downtime = true;
822 info->expected_downtime = s->expected_downtime;
823 info->has_setup_time = true;
824 info->setup_time = s->setup_time;
826 populate_ram_info(info, s);
827 populate_disk_info(info);
828 break;
829 case MIGRATION_STATUS_COLO:
830 info->has_status = true;
831 /* TODO: display COLO specific information (checkpoint info etc.) */
832 break;
833 case MIGRATION_STATUS_COMPLETED:
834 info->has_status = true;
835 info->has_total_time = true;
836 info->total_time = s->total_time;
837 info->has_downtime = true;
838 info->downtime = s->downtime;
839 info->has_setup_time = true;
840 info->setup_time = s->setup_time;
842 populate_ram_info(info, s);
843 break;
844 case MIGRATION_STATUS_FAILED:
845 info->has_status = true;
846 if (s->error) {
847 info->has_error_desc = true;
848 info->error_desc = g_strdup(error_get_pretty(s->error));
850 break;
851 case MIGRATION_STATUS_CANCELLED:
852 info->has_status = true;
853 break;
855 info->status = s->state;
859 * @migration_caps_check - check capability validity
861 * @cap_list: old capability list, array of bool
862 * @params: new capabilities to be applied soon
863 * @errp: set *errp if the check failed, with reason
865 * Returns true if check passed, otherwise false.
867 static bool migrate_caps_check(bool *cap_list,
868 MigrationCapabilityStatusList *params,
869 Error **errp)
871 MigrationCapabilityStatusList *cap;
872 bool old_postcopy_cap;
873 MigrationIncomingState *mis = migration_incoming_get_current();
875 old_postcopy_cap = cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM];
877 for (cap = params; cap; cap = cap->next) {
878 cap_list[cap->value->capability] = cap->value->state;
881 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
882 if (cap_list[MIGRATION_CAPABILITY_BLOCK]) {
883 error_setg(errp, "QEMU compiled without old-style (blk/-b, inc/-i) "
884 "block migration");
885 error_append_hint(errp, "Use drive_mirror+NBD instead.\n");
886 return false;
888 #endif
890 if (cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]) {
891 if (cap_list[MIGRATION_CAPABILITY_COMPRESS]) {
892 /* The decompression threads asynchronously write into RAM
893 * rather than use the atomic copies needed to avoid
894 * userfaulting. It should be possible to fix the decompression
895 * threads for compatibility in future.
897 error_setg(errp, "Postcopy is not currently compatible "
898 "with compression");
899 return false;
902 /* This check is reasonably expensive, so only when it's being
903 * set the first time, also it's only the destination that needs
904 * special support.
906 if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) &&
907 !postcopy_ram_supported_by_host(mis)) {
908 /* postcopy_ram_supported_by_host will have emitted a more
909 * detailed message
911 error_setg(errp, "Postcopy is not supported");
912 return false;
916 return true;
919 static void fill_destination_migration_info(MigrationInfo *info)
921 MigrationIncomingState *mis = migration_incoming_get_current();
923 switch (mis->state) {
924 case MIGRATION_STATUS_NONE:
925 return;
926 break;
927 case MIGRATION_STATUS_SETUP:
928 case MIGRATION_STATUS_CANCELLING:
929 case MIGRATION_STATUS_CANCELLED:
930 case MIGRATION_STATUS_ACTIVE:
931 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
932 case MIGRATION_STATUS_POSTCOPY_PAUSED:
933 case MIGRATION_STATUS_POSTCOPY_RECOVER:
934 case MIGRATION_STATUS_FAILED:
935 case MIGRATION_STATUS_COLO:
936 info->has_status = true;
937 break;
938 case MIGRATION_STATUS_COMPLETED:
939 info->has_status = true;
940 fill_destination_postcopy_migration_info(info);
941 break;
943 info->status = mis->state;
946 MigrationInfo *qmp_query_migrate(Error **errp)
948 MigrationInfo *info = g_malloc0(sizeof(*info));
950 fill_destination_migration_info(info);
951 fill_source_migration_info(info);
953 return info;
956 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
957 Error **errp)
959 MigrationState *s = migrate_get_current();
960 MigrationCapabilityStatusList *cap;
961 bool cap_list[MIGRATION_CAPABILITY__MAX];
963 if (migration_is_setup_or_active(s->state)) {
964 error_setg(errp, QERR_MIGRATION_ACTIVE);
965 return;
968 memcpy(cap_list, s->enabled_capabilities, sizeof(cap_list));
969 if (!migrate_caps_check(cap_list, params, errp)) {
970 return;
973 for (cap = params; cap; cap = cap->next) {
974 s->enabled_capabilities[cap->value->capability] = cap->value->state;
979 * Check whether the parameters are valid. Error will be put into errp
980 * (if provided). Return true if valid, otherwise false.
982 static bool migrate_params_check(MigrationParameters *params, Error **errp)
984 if (params->has_compress_level &&
985 (params->compress_level > 9)) {
986 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
987 "is invalid, it should be in the range of 0 to 9");
988 return false;
991 if (params->has_compress_threads && (params->compress_threads < 1)) {
992 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
993 "compress_threads",
994 "is invalid, it should be in the range of 1 to 255");
995 return false;
998 if (params->has_decompress_threads && (params->decompress_threads < 1)) {
999 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1000 "decompress_threads",
1001 "is invalid, it should be in the range of 1 to 255");
1002 return false;
1005 if (params->has_cpu_throttle_initial &&
1006 (params->cpu_throttle_initial < 1 ||
1007 params->cpu_throttle_initial > 99)) {
1008 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1009 "cpu_throttle_initial",
1010 "an integer in the range of 1 to 99");
1011 return false;
1014 if (params->has_cpu_throttle_increment &&
1015 (params->cpu_throttle_increment < 1 ||
1016 params->cpu_throttle_increment > 99)) {
1017 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1018 "cpu_throttle_increment",
1019 "an integer in the range of 1 to 99");
1020 return false;
1023 if (params->has_max_bandwidth && (params->max_bandwidth > SIZE_MAX)) {
1024 error_setg(errp, "Parameter 'max_bandwidth' expects an integer in the"
1025 " range of 0 to %zu bytes/second", SIZE_MAX);
1026 return false;
1029 if (params->has_downtime_limit &&
1030 (params->downtime_limit > MAX_MIGRATE_DOWNTIME)) {
1031 error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
1032 "the range of 0 to %d milliseconds",
1033 MAX_MIGRATE_DOWNTIME);
1034 return false;
1037 /* x_checkpoint_delay is now always positive */
1039 if (params->has_x_multifd_channels && (params->x_multifd_channels < 1)) {
1040 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1041 "multifd_channels",
1042 "is invalid, it should be in the range of 1 to 255");
1043 return false;
1045 if (params->has_x_multifd_page_count &&
1046 (params->x_multifd_page_count < 1 ||
1047 params->x_multifd_page_count > 10000)) {
1048 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1049 "multifd_page_count",
1050 "is invalid, it should be in the range of 1 to 10000");
1051 return false;
1054 if (params->has_xbzrle_cache_size &&
1055 (params->xbzrle_cache_size < qemu_target_page_size() ||
1056 !is_power_of_2(params->xbzrle_cache_size))) {
1057 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1058 "xbzrle_cache_size",
1059 "is invalid, it should be bigger than target page size"
1060 " and a power of two");
1061 return false;
1064 if (params->has_max_cpu_throttle &&
1065 (params->max_cpu_throttle < params->cpu_throttle_initial ||
1066 params->max_cpu_throttle > 99)) {
1067 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1068 "max_cpu_throttle",
1069 "an integer in the range of cpu_throttle_initial to 99");
1070 return false;
1073 return true;
1076 static void migrate_params_test_apply(MigrateSetParameters *params,
1077 MigrationParameters *dest)
1079 *dest = migrate_get_current()->parameters;
1081 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1083 if (params->has_compress_level) {
1084 dest->compress_level = params->compress_level;
1087 if (params->has_compress_threads) {
1088 dest->compress_threads = params->compress_threads;
1091 if (params->has_compress_wait_thread) {
1092 dest->compress_wait_thread = params->compress_wait_thread;
1095 if (params->has_decompress_threads) {
1096 dest->decompress_threads = params->decompress_threads;
1099 if (params->has_cpu_throttle_initial) {
1100 dest->cpu_throttle_initial = params->cpu_throttle_initial;
1103 if (params->has_cpu_throttle_increment) {
1104 dest->cpu_throttle_increment = params->cpu_throttle_increment;
1107 if (params->has_tls_creds) {
1108 assert(params->tls_creds->type == QTYPE_QSTRING);
1109 dest->tls_creds = g_strdup(params->tls_creds->u.s);
1112 if (params->has_tls_hostname) {
1113 assert(params->tls_hostname->type == QTYPE_QSTRING);
1114 dest->tls_hostname = g_strdup(params->tls_hostname->u.s);
1117 if (params->has_max_bandwidth) {
1118 dest->max_bandwidth = params->max_bandwidth;
1121 if (params->has_downtime_limit) {
1122 dest->downtime_limit = params->downtime_limit;
1125 if (params->has_x_checkpoint_delay) {
1126 dest->x_checkpoint_delay = params->x_checkpoint_delay;
1129 if (params->has_block_incremental) {
1130 dest->block_incremental = params->block_incremental;
1132 if (params->has_x_multifd_channels) {
1133 dest->x_multifd_channels = params->x_multifd_channels;
1135 if (params->has_x_multifd_page_count) {
1136 dest->x_multifd_page_count = params->x_multifd_page_count;
1138 if (params->has_xbzrle_cache_size) {
1139 dest->xbzrle_cache_size = params->xbzrle_cache_size;
1141 if (params->has_max_postcopy_bandwidth) {
1142 dest->max_postcopy_bandwidth = params->max_postcopy_bandwidth;
1144 if (params->has_max_cpu_throttle) {
1145 dest->max_cpu_throttle = params->max_cpu_throttle;
1149 static void migrate_params_apply(MigrateSetParameters *params, Error **errp)
1151 MigrationState *s = migrate_get_current();
1153 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1155 if (params->has_compress_level) {
1156 s->parameters.compress_level = params->compress_level;
1159 if (params->has_compress_threads) {
1160 s->parameters.compress_threads = params->compress_threads;
1163 if (params->has_compress_wait_thread) {
1164 s->parameters.compress_wait_thread = params->compress_wait_thread;
1167 if (params->has_decompress_threads) {
1168 s->parameters.decompress_threads = params->decompress_threads;
1171 if (params->has_cpu_throttle_initial) {
1172 s->parameters.cpu_throttle_initial = params->cpu_throttle_initial;
1175 if (params->has_cpu_throttle_increment) {
1176 s->parameters.cpu_throttle_increment = params->cpu_throttle_increment;
1179 if (params->has_tls_creds) {
1180 g_free(s->parameters.tls_creds);
1181 assert(params->tls_creds->type == QTYPE_QSTRING);
1182 s->parameters.tls_creds = g_strdup(params->tls_creds->u.s);
1185 if (params->has_tls_hostname) {
1186 g_free(s->parameters.tls_hostname);
1187 assert(params->tls_hostname->type == QTYPE_QSTRING);
1188 s->parameters.tls_hostname = g_strdup(params->tls_hostname->u.s);
1191 if (params->has_max_bandwidth) {
1192 s->parameters.max_bandwidth = params->max_bandwidth;
1193 if (s->to_dst_file) {
1194 qemu_file_set_rate_limit(s->to_dst_file,
1195 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
1199 if (params->has_downtime_limit) {
1200 s->parameters.downtime_limit = params->downtime_limit;
1203 if (params->has_x_checkpoint_delay) {
1204 s->parameters.x_checkpoint_delay = params->x_checkpoint_delay;
1205 if (migration_in_colo_state()) {
1206 colo_checkpoint_notify(s);
1210 if (params->has_block_incremental) {
1211 s->parameters.block_incremental = params->block_incremental;
1213 if (params->has_x_multifd_channels) {
1214 s->parameters.x_multifd_channels = params->x_multifd_channels;
1216 if (params->has_x_multifd_page_count) {
1217 s->parameters.x_multifd_page_count = params->x_multifd_page_count;
1219 if (params->has_xbzrle_cache_size) {
1220 s->parameters.xbzrle_cache_size = params->xbzrle_cache_size;
1221 xbzrle_cache_resize(params->xbzrle_cache_size, errp);
1223 if (params->has_max_postcopy_bandwidth) {
1224 s->parameters.max_postcopy_bandwidth = params->max_postcopy_bandwidth;
1226 if (params->has_max_cpu_throttle) {
1227 s->parameters.max_cpu_throttle = params->max_cpu_throttle;
1231 void qmp_migrate_set_parameters(MigrateSetParameters *params, Error **errp)
1233 MigrationParameters tmp;
1235 /* TODO Rewrite "" to null instead */
1236 if (params->has_tls_creds
1237 && params->tls_creds->type == QTYPE_QNULL) {
1238 qobject_unref(params->tls_creds->u.n);
1239 params->tls_creds->type = QTYPE_QSTRING;
1240 params->tls_creds->u.s = strdup("");
1242 /* TODO Rewrite "" to null instead */
1243 if (params->has_tls_hostname
1244 && params->tls_hostname->type == QTYPE_QNULL) {
1245 qobject_unref(params->tls_hostname->u.n);
1246 params->tls_hostname->type = QTYPE_QSTRING;
1247 params->tls_hostname->u.s = strdup("");
1250 migrate_params_test_apply(params, &tmp);
1252 if (!migrate_params_check(&tmp, errp)) {
1253 /* Invalid parameter */
1254 return;
1257 migrate_params_apply(params, errp);
1261 void qmp_migrate_start_postcopy(Error **errp)
1263 MigrationState *s = migrate_get_current();
1265 if (!migrate_postcopy()) {
1266 error_setg(errp, "Enable postcopy with migrate_set_capability before"
1267 " the start of migration");
1268 return;
1271 if (s->state == MIGRATION_STATUS_NONE) {
1272 error_setg(errp, "Postcopy must be started after migration has been"
1273 " started");
1274 return;
1277 * we don't error if migration has finished since that would be racy
1278 * with issuing this command.
1280 atomic_set(&s->start_postcopy, true);
1283 /* shared migration helpers */
1285 void migrate_set_state(int *state, int old_state, int new_state)
1287 assert(new_state < MIGRATION_STATUS__MAX);
1288 if (atomic_cmpxchg(state, old_state, new_state) == old_state) {
1289 trace_migrate_set_state(MigrationStatus_str(new_state));
1290 migrate_generate_event(new_state);
1294 static MigrationCapabilityStatusList *migrate_cap_add(
1295 MigrationCapabilityStatusList *list,
1296 MigrationCapability index,
1297 bool state)
1299 MigrationCapabilityStatusList *cap;
1301 cap = g_new0(MigrationCapabilityStatusList, 1);
1302 cap->value = g_new0(MigrationCapabilityStatus, 1);
1303 cap->value->capability = index;
1304 cap->value->state = state;
1305 cap->next = list;
1307 return cap;
1310 void migrate_set_block_enabled(bool value, Error **errp)
1312 MigrationCapabilityStatusList *cap;
1314 cap = migrate_cap_add(NULL, MIGRATION_CAPABILITY_BLOCK, value);
1315 qmp_migrate_set_capabilities(cap, errp);
1316 qapi_free_MigrationCapabilityStatusList(cap);
1319 static void migrate_set_block_incremental(MigrationState *s, bool value)
1321 s->parameters.block_incremental = value;
1324 static void block_cleanup_parameters(MigrationState *s)
1326 if (s->must_remove_block_options) {
1327 /* setting to false can never fail */
1328 migrate_set_block_enabled(false, &error_abort);
1329 migrate_set_block_incremental(s, false);
1330 s->must_remove_block_options = false;
1334 static void migrate_fd_cleanup(void *opaque)
1336 MigrationState *s = opaque;
1338 qemu_bh_delete(s->cleanup_bh);
1339 s->cleanup_bh = NULL;
1341 qemu_savevm_state_cleanup();
1343 if (s->to_dst_file) {
1344 Error *local_err = NULL;
1345 QEMUFile *tmp;
1347 trace_migrate_fd_cleanup();
1348 qemu_mutex_unlock_iothread();
1349 if (s->migration_thread_running) {
1350 qemu_thread_join(&s->thread);
1351 s->migration_thread_running = false;
1353 qemu_mutex_lock_iothread();
1355 if (multifd_save_cleanup(&local_err) != 0) {
1356 error_report_err(local_err);
1358 qemu_mutex_lock(&s->qemu_file_lock);
1359 tmp = s->to_dst_file;
1360 s->to_dst_file = NULL;
1361 qemu_mutex_unlock(&s->qemu_file_lock);
1363 * Close the file handle without the lock to make sure the
1364 * critical section won't block for long.
1366 qemu_fclose(tmp);
1369 assert((s->state != MIGRATION_STATUS_ACTIVE) &&
1370 (s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE));
1372 if (s->state == MIGRATION_STATUS_CANCELLING) {
1373 migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING,
1374 MIGRATION_STATUS_CANCELLED);
1377 if (s->error) {
1378 /* It is used on info migrate. We can't free it */
1379 error_report_err(error_copy(s->error));
1381 notifier_list_notify(&migration_state_notifiers, s);
1382 block_cleanup_parameters(s);
1385 void migrate_set_error(MigrationState *s, const Error *error)
1387 qemu_mutex_lock(&s->error_mutex);
1388 if (!s->error) {
1389 s->error = error_copy(error);
1391 qemu_mutex_unlock(&s->error_mutex);
1394 void migrate_fd_error(MigrationState *s, const Error *error)
1396 trace_migrate_fd_error(error_get_pretty(error));
1397 assert(s->to_dst_file == NULL);
1398 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1399 MIGRATION_STATUS_FAILED);
1400 migrate_set_error(s, error);
1403 static void migrate_fd_cancel(MigrationState *s)
1405 int old_state ;
1406 QEMUFile *f = migrate_get_current()->to_dst_file;
1407 trace_migrate_fd_cancel();
1409 if (s->rp_state.from_dst_file) {
1410 /* shutdown the rp socket, so causing the rp thread to shutdown */
1411 qemu_file_shutdown(s->rp_state.from_dst_file);
1414 do {
1415 old_state = s->state;
1416 if (!migration_is_setup_or_active(old_state)) {
1417 break;
1419 /* If the migration is paused, kick it out of the pause */
1420 if (old_state == MIGRATION_STATUS_PRE_SWITCHOVER) {
1421 qemu_sem_post(&s->pause_sem);
1423 migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING);
1424 } while (s->state != MIGRATION_STATUS_CANCELLING);
1427 * If we're unlucky the migration code might be stuck somewhere in a
1428 * send/write while the network has failed and is waiting to timeout;
1429 * if we've got shutdown(2) available then we can force it to quit.
1430 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1431 * called in a bh, so there is no race against this cancel.
1433 if (s->state == MIGRATION_STATUS_CANCELLING && f) {
1434 qemu_file_shutdown(f);
1436 if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) {
1437 Error *local_err = NULL;
1439 bdrv_invalidate_cache_all(&local_err);
1440 if (local_err) {
1441 error_report_err(local_err);
1442 } else {
1443 s->block_inactive = false;
1448 void add_migration_state_change_notifier(Notifier *notify)
1450 notifier_list_add(&migration_state_notifiers, notify);
1453 void remove_migration_state_change_notifier(Notifier *notify)
1455 notifier_remove(notify);
1458 bool migration_in_setup(MigrationState *s)
1460 return s->state == MIGRATION_STATUS_SETUP;
1463 bool migration_has_finished(MigrationState *s)
1465 return s->state == MIGRATION_STATUS_COMPLETED;
1468 bool migration_has_failed(MigrationState *s)
1470 return (s->state == MIGRATION_STATUS_CANCELLED ||
1471 s->state == MIGRATION_STATUS_FAILED);
1474 bool migration_in_postcopy(void)
1476 MigrationState *s = migrate_get_current();
1478 return (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
1481 bool migration_in_postcopy_after_devices(MigrationState *s)
1483 return migration_in_postcopy() && s->postcopy_after_devices;
1486 bool migration_is_idle(void)
1488 MigrationState *s = migrate_get_current();
1490 switch (s->state) {
1491 case MIGRATION_STATUS_NONE:
1492 case MIGRATION_STATUS_CANCELLED:
1493 case MIGRATION_STATUS_COMPLETED:
1494 case MIGRATION_STATUS_FAILED:
1495 return true;
1496 case MIGRATION_STATUS_SETUP:
1497 case MIGRATION_STATUS_CANCELLING:
1498 case MIGRATION_STATUS_ACTIVE:
1499 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1500 case MIGRATION_STATUS_COLO:
1501 case MIGRATION_STATUS_PRE_SWITCHOVER:
1502 case MIGRATION_STATUS_DEVICE:
1503 return false;
1504 case MIGRATION_STATUS__MAX:
1505 g_assert_not_reached();
1508 return false;
1511 void migrate_init(MigrationState *s)
1514 * Reinitialise all migration state, except
1515 * parameters/capabilities that the user set, and
1516 * locks.
1518 s->bytes_xfer = 0;
1519 s->xfer_limit = 0;
1520 s->cleanup_bh = 0;
1521 s->to_dst_file = NULL;
1522 s->state = MIGRATION_STATUS_NONE;
1523 s->rp_state.from_dst_file = NULL;
1524 s->rp_state.error = false;
1525 s->mbps = 0.0;
1526 s->downtime = 0;
1527 s->expected_downtime = 0;
1528 s->setup_time = 0;
1529 s->start_postcopy = false;
1530 s->postcopy_after_devices = false;
1531 s->migration_thread_running = false;
1532 error_free(s->error);
1533 s->error = NULL;
1535 migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
1537 s->start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1538 s->total_time = 0;
1539 s->vm_was_running = false;
1540 s->iteration_initial_bytes = 0;
1541 s->threshold_size = 0;
1544 static GSList *migration_blockers;
1546 int migrate_add_blocker(Error *reason, Error **errp)
1548 if (migrate_get_current()->only_migratable) {
1549 error_propagate(errp, error_copy(reason));
1550 error_prepend(errp, "disallowing migration blocker "
1551 "(--only_migratable) for: ");
1552 return -EACCES;
1555 if (migration_is_idle()) {
1556 migration_blockers = g_slist_prepend(migration_blockers, reason);
1557 return 0;
1560 error_propagate(errp, error_copy(reason));
1561 error_prepend(errp, "disallowing migration blocker (migration in "
1562 "progress) for: ");
1563 return -EBUSY;
1566 void migrate_del_blocker(Error *reason)
1568 migration_blockers = g_slist_remove(migration_blockers, reason);
1571 void qmp_migrate_incoming(const char *uri, Error **errp)
1573 Error *local_err = NULL;
1574 static bool once = true;
1576 if (!deferred_incoming) {
1577 error_setg(errp, "For use with '-incoming defer'");
1578 return;
1580 if (!once) {
1581 error_setg(errp, "The incoming migration has already been started");
1584 qemu_start_incoming_migration(uri, &local_err);
1586 if (local_err) {
1587 error_propagate(errp, local_err);
1588 return;
1591 once = false;
1594 void qmp_migrate_recover(const char *uri, Error **errp)
1596 MigrationIncomingState *mis = migration_incoming_get_current();
1598 if (mis->state != MIGRATION_STATUS_POSTCOPY_PAUSED) {
1599 error_setg(errp, "Migrate recover can only be run "
1600 "when postcopy is paused.");
1601 return;
1604 if (atomic_cmpxchg(&mis->postcopy_recover_triggered,
1605 false, true) == true) {
1606 error_setg(errp, "Migrate recovery is triggered already");
1607 return;
1611 * Note that this call will never start a real migration; it will
1612 * only re-setup the migration stream and poke existing migration
1613 * to continue using that newly established channel.
1615 qemu_start_incoming_migration(uri, errp);
1618 void qmp_migrate_pause(Error **errp)
1620 MigrationState *ms = migrate_get_current();
1621 MigrationIncomingState *mis = migration_incoming_get_current();
1622 int ret;
1624 if (ms->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
1625 /* Source side, during postcopy */
1626 qemu_mutex_lock(&ms->qemu_file_lock);
1627 ret = qemu_file_shutdown(ms->to_dst_file);
1628 qemu_mutex_unlock(&ms->qemu_file_lock);
1629 if (ret) {
1630 error_setg(errp, "Failed to pause source migration");
1632 return;
1635 if (mis->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
1636 ret = qemu_file_shutdown(mis->from_src_file);
1637 if (ret) {
1638 error_setg(errp, "Failed to pause destination migration");
1640 return;
1643 error_setg(errp, "migrate-pause is currently only supported "
1644 "during postcopy-active state");
1647 bool migration_is_blocked(Error **errp)
1649 if (qemu_savevm_state_blocked(errp)) {
1650 return true;
1653 if (migration_blockers) {
1654 error_propagate(errp, error_copy(migration_blockers->data));
1655 return true;
1658 return false;
1661 /* Returns true if continue to migrate, or false if error detected */
1662 static bool migrate_prepare(MigrationState *s, bool blk, bool blk_inc,
1663 bool resume, Error **errp)
1665 Error *local_err = NULL;
1667 if (resume) {
1668 if (s->state != MIGRATION_STATUS_POSTCOPY_PAUSED) {
1669 error_setg(errp, "Cannot resume if there is no "
1670 "paused migration");
1671 return false;
1675 * Postcopy recovery won't work well with release-ram
1676 * capability since release-ram will drop the page buffer as
1677 * long as the page is put into the send buffer. So if there
1678 * is a network failure happened, any page buffers that have
1679 * not yet reached the destination VM but have already been
1680 * sent from the source VM will be lost forever. Let's refuse
1681 * the client from resuming such a postcopy migration.
1682 * Luckily release-ram was designed to only be used when src
1683 * and destination VMs are on the same host, so it should be
1684 * fine.
1686 if (migrate_release_ram()) {
1687 error_setg(errp, "Postcopy recovery cannot work "
1688 "when release-ram capability is set");
1689 return false;
1692 /* This is a resume, skip init status */
1693 return true;
1696 if (migration_is_setup_or_active(s->state) ||
1697 s->state == MIGRATION_STATUS_CANCELLING ||
1698 s->state == MIGRATION_STATUS_COLO) {
1699 error_setg(errp, QERR_MIGRATION_ACTIVE);
1700 return false;
1703 if (runstate_check(RUN_STATE_INMIGRATE)) {
1704 error_setg(errp, "Guest is waiting for an incoming migration");
1705 return false;
1708 if (migration_is_blocked(errp)) {
1709 return false;
1712 if (blk || blk_inc) {
1713 if (migrate_use_block() || migrate_use_block_incremental()) {
1714 error_setg(errp, "Command options are incompatible with "
1715 "current migration capabilities");
1716 return false;
1718 migrate_set_block_enabled(true, &local_err);
1719 if (local_err) {
1720 error_propagate(errp, local_err);
1721 return false;
1723 s->must_remove_block_options = true;
1726 if (blk_inc) {
1727 migrate_set_block_incremental(s, true);
1730 migrate_init(s);
1732 return true;
1735 void qmp_migrate(const char *uri, bool has_blk, bool blk,
1736 bool has_inc, bool inc, bool has_detach, bool detach,
1737 bool has_resume, bool resume, Error **errp)
1739 Error *local_err = NULL;
1740 MigrationState *s = migrate_get_current();
1741 const char *p;
1743 if (!migrate_prepare(s, has_blk && blk, has_inc && inc,
1744 has_resume && resume, errp)) {
1745 /* Error detected, put into errp */
1746 return;
1749 if (strstart(uri, "tcp:", &p)) {
1750 tcp_start_outgoing_migration(s, p, &local_err);
1751 #ifdef CONFIG_RDMA
1752 } else if (strstart(uri, "rdma:", &p)) {
1753 rdma_start_outgoing_migration(s, p, &local_err);
1754 #endif
1755 } else if (strstart(uri, "exec:", &p)) {
1756 exec_start_outgoing_migration(s, p, &local_err);
1757 } else if (strstart(uri, "unix:", &p)) {
1758 unix_start_outgoing_migration(s, p, &local_err);
1759 } else if (strstart(uri, "fd:", &p)) {
1760 fd_start_outgoing_migration(s, p, &local_err);
1761 } else {
1762 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
1763 "a valid migration protocol");
1764 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1765 MIGRATION_STATUS_FAILED);
1766 block_cleanup_parameters(s);
1767 return;
1770 if (local_err) {
1771 migrate_fd_error(s, local_err);
1772 error_propagate(errp, local_err);
1773 return;
1777 void qmp_migrate_cancel(Error **errp)
1779 migrate_fd_cancel(migrate_get_current());
1782 void qmp_migrate_continue(MigrationStatus state, Error **errp)
1784 MigrationState *s = migrate_get_current();
1785 if (s->state != state) {
1786 error_setg(errp, "Migration not in expected state: %s",
1787 MigrationStatus_str(s->state));
1788 return;
1790 qemu_sem_post(&s->pause_sem);
1793 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
1795 MigrateSetParameters p = {
1796 .has_xbzrle_cache_size = true,
1797 .xbzrle_cache_size = value,
1800 qmp_migrate_set_parameters(&p, errp);
1803 int64_t qmp_query_migrate_cache_size(Error **errp)
1805 return migrate_xbzrle_cache_size();
1808 void qmp_migrate_set_speed(int64_t value, Error **errp)
1810 MigrateSetParameters p = {
1811 .has_max_bandwidth = true,
1812 .max_bandwidth = value,
1815 qmp_migrate_set_parameters(&p, errp);
1818 void qmp_migrate_set_downtime(double value, Error **errp)
1820 if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) {
1821 error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
1822 "the range of 0 to %d seconds",
1823 MAX_MIGRATE_DOWNTIME_SECONDS);
1824 return;
1827 value *= 1000; /* Convert to milliseconds */
1828 value = MAX(0, MIN(INT64_MAX, value));
1830 MigrateSetParameters p = {
1831 .has_downtime_limit = true,
1832 .downtime_limit = value,
1835 qmp_migrate_set_parameters(&p, errp);
1838 bool migrate_release_ram(void)
1840 MigrationState *s;
1842 s = migrate_get_current();
1844 return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM];
1847 bool migrate_postcopy_ram(void)
1849 MigrationState *s;
1851 s = migrate_get_current();
1853 return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM];
1856 bool migrate_postcopy(void)
1858 return migrate_postcopy_ram() || migrate_dirty_bitmaps();
1861 bool migrate_auto_converge(void)
1863 MigrationState *s;
1865 s = migrate_get_current();
1867 return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
1870 bool migrate_zero_blocks(void)
1872 MigrationState *s;
1874 s = migrate_get_current();
1876 return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
1879 bool migrate_postcopy_blocktime(void)
1881 MigrationState *s;
1883 s = migrate_get_current();
1885 return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME];
1888 bool migrate_use_compression(void)
1890 MigrationState *s;
1892 s = migrate_get_current();
1894 return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
1897 int migrate_compress_level(void)
1899 MigrationState *s;
1901 s = migrate_get_current();
1903 return s->parameters.compress_level;
1906 int migrate_compress_threads(void)
1908 MigrationState *s;
1910 s = migrate_get_current();
1912 return s->parameters.compress_threads;
1915 int migrate_compress_wait_thread(void)
1917 MigrationState *s;
1919 s = migrate_get_current();
1921 return s->parameters.compress_wait_thread;
1924 int migrate_decompress_threads(void)
1926 MigrationState *s;
1928 s = migrate_get_current();
1930 return s->parameters.decompress_threads;
1933 bool migrate_dirty_bitmaps(void)
1935 MigrationState *s;
1937 s = migrate_get_current();
1939 return s->enabled_capabilities[MIGRATION_CAPABILITY_DIRTY_BITMAPS];
1942 bool migrate_use_events(void)
1944 MigrationState *s;
1946 s = migrate_get_current();
1948 return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
1951 bool migrate_use_multifd(void)
1953 MigrationState *s;
1955 s = migrate_get_current();
1957 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_MULTIFD];
1960 bool migrate_pause_before_switchover(void)
1962 MigrationState *s;
1964 s = migrate_get_current();
1966 return s->enabled_capabilities[
1967 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER];
1970 int migrate_multifd_channels(void)
1972 MigrationState *s;
1974 s = migrate_get_current();
1976 return s->parameters.x_multifd_channels;
1979 int migrate_multifd_page_count(void)
1981 MigrationState *s;
1983 s = migrate_get_current();
1985 return s->parameters.x_multifd_page_count;
1988 int migrate_use_xbzrle(void)
1990 MigrationState *s;
1992 s = migrate_get_current();
1994 return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
1997 int64_t migrate_xbzrle_cache_size(void)
1999 MigrationState *s;
2001 s = migrate_get_current();
2003 return s->parameters.xbzrle_cache_size;
2006 static int64_t migrate_max_postcopy_bandwidth(void)
2008 MigrationState *s;
2010 s = migrate_get_current();
2012 return s->parameters.max_postcopy_bandwidth;
2015 bool migrate_use_block(void)
2017 MigrationState *s;
2019 s = migrate_get_current();
2021 return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK];
2024 bool migrate_use_return_path(void)
2026 MigrationState *s;
2028 s = migrate_get_current();
2030 return s->enabled_capabilities[MIGRATION_CAPABILITY_RETURN_PATH];
2033 bool migrate_use_block_incremental(void)
2035 MigrationState *s;
2037 s = migrate_get_current();
2039 return s->parameters.block_incremental;
2042 /* migration thread support */
2044 * Something bad happened to the RP stream, mark an error
2045 * The caller shall print or trace something to indicate why
2047 static void mark_source_rp_bad(MigrationState *s)
2049 s->rp_state.error = true;
2052 static struct rp_cmd_args {
2053 ssize_t len; /* -1 = variable */
2054 const char *name;
2055 } rp_cmd_args[] = {
2056 [MIG_RP_MSG_INVALID] = { .len = -1, .name = "INVALID" },
2057 [MIG_RP_MSG_SHUT] = { .len = 4, .name = "SHUT" },
2058 [MIG_RP_MSG_PONG] = { .len = 4, .name = "PONG" },
2059 [MIG_RP_MSG_REQ_PAGES] = { .len = 12, .name = "REQ_PAGES" },
2060 [MIG_RP_MSG_REQ_PAGES_ID] = { .len = -1, .name = "REQ_PAGES_ID" },
2061 [MIG_RP_MSG_RECV_BITMAP] = { .len = -1, .name = "RECV_BITMAP" },
2062 [MIG_RP_MSG_RESUME_ACK] = { .len = 4, .name = "RESUME_ACK" },
2063 [MIG_RP_MSG_MAX] = { .len = -1, .name = "MAX" },
2067 * Process a request for pages received on the return path,
2068 * We're allowed to send more than requested (e.g. to round to our page size)
2069 * and we don't need to send pages that have already been sent.
2071 static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname,
2072 ram_addr_t start, size_t len)
2074 long our_host_ps = getpagesize();
2076 trace_migrate_handle_rp_req_pages(rbname, start, len);
2079 * Since we currently insist on matching page sizes, just sanity check
2080 * we're being asked for whole host pages.
2082 if (start & (our_host_ps-1) ||
2083 (len & (our_host_ps-1))) {
2084 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
2085 " len: %zd", __func__, start, len);
2086 mark_source_rp_bad(ms);
2087 return;
2090 if (ram_save_queue_pages(rbname, start, len)) {
2091 mark_source_rp_bad(ms);
2095 /* Return true to retry, false to quit */
2096 static bool postcopy_pause_return_path_thread(MigrationState *s)
2098 trace_postcopy_pause_return_path();
2100 qemu_sem_wait(&s->postcopy_pause_rp_sem);
2102 trace_postcopy_pause_return_path_continued();
2104 return true;
2107 static int migrate_handle_rp_recv_bitmap(MigrationState *s, char *block_name)
2109 RAMBlock *block = qemu_ram_block_by_name(block_name);
2111 if (!block) {
2112 error_report("%s: invalid block name '%s'", __func__, block_name);
2113 return -EINVAL;
2116 /* Fetch the received bitmap and refresh the dirty bitmap */
2117 return ram_dirty_bitmap_reload(s, block);
2120 static int migrate_handle_rp_resume_ack(MigrationState *s, uint32_t value)
2122 trace_source_return_path_thread_resume_ack(value);
2124 if (value != MIGRATION_RESUME_ACK_VALUE) {
2125 error_report("%s: illegal resume_ack value %"PRIu32,
2126 __func__, value);
2127 return -1;
2130 /* Now both sides are active. */
2131 migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_RECOVER,
2132 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2134 /* Notify send thread that time to continue send pages */
2135 qemu_sem_post(&s->rp_state.rp_sem);
2137 return 0;
2141 * Handles messages sent on the return path towards the source VM
2144 static void *source_return_path_thread(void *opaque)
2146 MigrationState *ms = opaque;
2147 QEMUFile *rp = ms->rp_state.from_dst_file;
2148 uint16_t header_len, header_type;
2149 uint8_t buf[512];
2150 uint32_t tmp32, sibling_error;
2151 ram_addr_t start = 0; /* =0 to silence warning */
2152 size_t len = 0, expected_len;
2153 int res;
2155 trace_source_return_path_thread_entry();
2156 rcu_register_thread();
2158 retry:
2159 while (!ms->rp_state.error && !qemu_file_get_error(rp) &&
2160 migration_is_setup_or_active(ms->state)) {
2161 trace_source_return_path_thread_loop_top();
2162 header_type = qemu_get_be16(rp);
2163 header_len = qemu_get_be16(rp);
2165 if (qemu_file_get_error(rp)) {
2166 mark_source_rp_bad(ms);
2167 goto out;
2170 if (header_type >= MIG_RP_MSG_MAX ||
2171 header_type == MIG_RP_MSG_INVALID) {
2172 error_report("RP: Received invalid message 0x%04x length 0x%04x",
2173 header_type, header_len);
2174 mark_source_rp_bad(ms);
2175 goto out;
2178 if ((rp_cmd_args[header_type].len != -1 &&
2179 header_len != rp_cmd_args[header_type].len) ||
2180 header_len > sizeof(buf)) {
2181 error_report("RP: Received '%s' message (0x%04x) with"
2182 "incorrect length %d expecting %zu",
2183 rp_cmd_args[header_type].name, header_type, header_len,
2184 (size_t)rp_cmd_args[header_type].len);
2185 mark_source_rp_bad(ms);
2186 goto out;
2189 /* We know we've got a valid header by this point */
2190 res = qemu_get_buffer(rp, buf, header_len);
2191 if (res != header_len) {
2192 error_report("RP: Failed reading data for message 0x%04x"
2193 " read %d expected %d",
2194 header_type, res, header_len);
2195 mark_source_rp_bad(ms);
2196 goto out;
2199 /* OK, we have the message and the data */
2200 switch (header_type) {
2201 case MIG_RP_MSG_SHUT:
2202 sibling_error = ldl_be_p(buf);
2203 trace_source_return_path_thread_shut(sibling_error);
2204 if (sibling_error) {
2205 error_report("RP: Sibling indicated error %d", sibling_error);
2206 mark_source_rp_bad(ms);
2209 * We'll let the main thread deal with closing the RP
2210 * we could do a shutdown(2) on it, but we're the only user
2211 * anyway, so there's nothing gained.
2213 goto out;
2215 case MIG_RP_MSG_PONG:
2216 tmp32 = ldl_be_p(buf);
2217 trace_source_return_path_thread_pong(tmp32);
2218 break;
2220 case MIG_RP_MSG_REQ_PAGES:
2221 start = ldq_be_p(buf);
2222 len = ldl_be_p(buf + 8);
2223 migrate_handle_rp_req_pages(ms, NULL, start, len);
2224 break;
2226 case MIG_RP_MSG_REQ_PAGES_ID:
2227 expected_len = 12 + 1; /* header + termination */
2229 if (header_len >= expected_len) {
2230 start = ldq_be_p(buf);
2231 len = ldl_be_p(buf + 8);
2232 /* Now we expect an idstr */
2233 tmp32 = buf[12]; /* Length of the following idstr */
2234 buf[13 + tmp32] = '\0';
2235 expected_len += tmp32;
2237 if (header_len != expected_len) {
2238 error_report("RP: Req_Page_id with length %d expecting %zd",
2239 header_len, expected_len);
2240 mark_source_rp_bad(ms);
2241 goto out;
2243 migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len);
2244 break;
2246 case MIG_RP_MSG_RECV_BITMAP:
2247 if (header_len < 1) {
2248 error_report("%s: missing block name", __func__);
2249 mark_source_rp_bad(ms);
2250 goto out;
2252 /* Format: len (1B) + idstr (<255B). This ends the idstr. */
2253 buf[buf[0] + 1] = '\0';
2254 if (migrate_handle_rp_recv_bitmap(ms, (char *)(buf + 1))) {
2255 mark_source_rp_bad(ms);
2256 goto out;
2258 break;
2260 case MIG_RP_MSG_RESUME_ACK:
2261 tmp32 = ldl_be_p(buf);
2262 if (migrate_handle_rp_resume_ack(ms, tmp32)) {
2263 mark_source_rp_bad(ms);
2264 goto out;
2266 break;
2268 default:
2269 break;
2273 out:
2274 res = qemu_file_get_error(rp);
2275 if (res) {
2276 if (res == -EIO) {
2278 * Maybe there is something we can do: it looks like a
2279 * network down issue, and we pause for a recovery.
2281 if (postcopy_pause_return_path_thread(ms)) {
2282 /* Reload rp, reset the rest */
2283 if (rp != ms->rp_state.from_dst_file) {
2284 qemu_fclose(rp);
2285 rp = ms->rp_state.from_dst_file;
2287 ms->rp_state.error = false;
2288 goto retry;
2292 trace_source_return_path_thread_bad_end();
2293 mark_source_rp_bad(ms);
2296 trace_source_return_path_thread_end();
2297 ms->rp_state.from_dst_file = NULL;
2298 qemu_fclose(rp);
2299 rcu_unregister_thread();
2300 return NULL;
2303 static int open_return_path_on_source(MigrationState *ms,
2304 bool create_thread)
2307 ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file);
2308 if (!ms->rp_state.from_dst_file) {
2309 return -1;
2312 trace_open_return_path_on_source();
2314 if (!create_thread) {
2315 /* We're done */
2316 return 0;
2319 qemu_thread_create(&ms->rp_state.rp_thread, "return path",
2320 source_return_path_thread, ms, QEMU_THREAD_JOINABLE);
2322 trace_open_return_path_on_source_continue();
2324 return 0;
2327 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
2328 static int await_return_path_close_on_source(MigrationState *ms)
2331 * If this is a normal exit then the destination will send a SHUT and the
2332 * rp_thread will exit, however if there's an error we need to cause
2333 * it to exit.
2335 if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) {
2337 * shutdown(2), if we have it, will cause it to unblock if it's stuck
2338 * waiting for the destination.
2340 qemu_file_shutdown(ms->rp_state.from_dst_file);
2341 mark_source_rp_bad(ms);
2343 trace_await_return_path_close_on_source_joining();
2344 qemu_thread_join(&ms->rp_state.rp_thread);
2345 trace_await_return_path_close_on_source_close();
2346 return ms->rp_state.error;
2350 * Switch from normal iteration to postcopy
2351 * Returns non-0 on error
2353 static int postcopy_start(MigrationState *ms)
2355 int ret;
2356 QIOChannelBuffer *bioc;
2357 QEMUFile *fb;
2358 int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2359 int64_t bandwidth = migrate_max_postcopy_bandwidth();
2360 bool restart_block = false;
2361 int cur_state = MIGRATION_STATUS_ACTIVE;
2362 if (!migrate_pause_before_switchover()) {
2363 migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE,
2364 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2367 trace_postcopy_start();
2368 qemu_mutex_lock_iothread();
2369 trace_postcopy_start_set_run();
2371 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
2372 global_state_store();
2373 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
2374 if (ret < 0) {
2375 goto fail;
2378 ret = migration_maybe_pause(ms, &cur_state,
2379 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2380 if (ret < 0) {
2381 goto fail;
2384 ret = bdrv_inactivate_all();
2385 if (ret < 0) {
2386 goto fail;
2388 restart_block = true;
2391 * Cause any non-postcopiable, but iterative devices to
2392 * send out their final data.
2394 qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false);
2397 * in Finish migrate and with the io-lock held everything should
2398 * be quiet, but we've potentially still got dirty pages and we
2399 * need to tell the destination to throw any pages it's already received
2400 * that are dirty
2402 if (migrate_postcopy_ram()) {
2403 if (ram_postcopy_send_discard_bitmap(ms)) {
2404 error_report("postcopy send discard bitmap failed");
2405 goto fail;
2410 * send rest of state - note things that are doing postcopy
2411 * will notice we're in POSTCOPY_ACTIVE and not actually
2412 * wrap their state up here
2414 /* 0 max-postcopy-bandwidth means unlimited */
2415 if (!bandwidth) {
2416 qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX);
2417 } else {
2418 qemu_file_set_rate_limit(ms->to_dst_file, bandwidth / XFER_LIMIT_RATIO);
2420 if (migrate_postcopy_ram()) {
2421 /* Ping just for debugging, helps line traces up */
2422 qemu_savevm_send_ping(ms->to_dst_file, 2);
2426 * While loading the device state we may trigger page transfer
2427 * requests and the fd must be free to process those, and thus
2428 * the destination must read the whole device state off the fd before
2429 * it starts processing it. Unfortunately the ad-hoc migration format
2430 * doesn't allow the destination to know the size to read without fully
2431 * parsing it through each devices load-state code (especially the open
2432 * coded devices that use get/put).
2433 * So we wrap the device state up in a package with a length at the start;
2434 * to do this we use a qemu_buf to hold the whole of the device state.
2436 bioc = qio_channel_buffer_new(4096);
2437 qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer");
2438 fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
2439 object_unref(OBJECT(bioc));
2442 * Make sure the receiver can get incoming pages before we send the rest
2443 * of the state
2445 qemu_savevm_send_postcopy_listen(fb);
2447 qemu_savevm_state_complete_precopy(fb, false, false);
2448 if (migrate_postcopy_ram()) {
2449 qemu_savevm_send_ping(fb, 3);
2452 qemu_savevm_send_postcopy_run(fb);
2454 /* <><> end of stuff going into the package */
2456 /* Last point of recovery; as soon as we send the package the destination
2457 * can open devices and potentially start running.
2458 * Lets just check again we've not got any errors.
2460 ret = qemu_file_get_error(ms->to_dst_file);
2461 if (ret) {
2462 error_report("postcopy_start: Migration stream errored (pre package)");
2463 goto fail_closefb;
2466 restart_block = false;
2468 /* Now send that blob */
2469 if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) {
2470 goto fail_closefb;
2472 qemu_fclose(fb);
2474 /* Send a notify to give a chance for anything that needs to happen
2475 * at the transition to postcopy and after the device state; in particular
2476 * spice needs to trigger a transition now
2478 ms->postcopy_after_devices = true;
2479 notifier_list_notify(&migration_state_notifiers, ms);
2481 ms->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop;
2483 qemu_mutex_unlock_iothread();
2485 if (migrate_postcopy_ram()) {
2487 * Although this ping is just for debug, it could potentially be
2488 * used for getting a better measurement of downtime at the source.
2490 qemu_savevm_send_ping(ms->to_dst_file, 4);
2493 if (migrate_release_ram()) {
2494 ram_postcopy_migrated_memory_release(ms);
2497 ret = qemu_file_get_error(ms->to_dst_file);
2498 if (ret) {
2499 error_report("postcopy_start: Migration stream errored");
2500 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
2501 MIGRATION_STATUS_FAILED);
2504 return ret;
2506 fail_closefb:
2507 qemu_fclose(fb);
2508 fail:
2509 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
2510 MIGRATION_STATUS_FAILED);
2511 if (restart_block) {
2512 /* A failure happened early enough that we know the destination hasn't
2513 * accessed block devices, so we're safe to recover.
2515 Error *local_err = NULL;
2517 bdrv_invalidate_cache_all(&local_err);
2518 if (local_err) {
2519 error_report_err(local_err);
2522 qemu_mutex_unlock_iothread();
2523 return -1;
2527 * migration_maybe_pause: Pause if required to by
2528 * migrate_pause_before_switchover called with the iothread locked
2529 * Returns: 0 on success
2531 static int migration_maybe_pause(MigrationState *s,
2532 int *current_active_state,
2533 int new_state)
2535 if (!migrate_pause_before_switchover()) {
2536 return 0;
2539 /* Since leaving this state is not atomic with posting the semaphore
2540 * it's possible that someone could have issued multiple migrate_continue
2541 * and the semaphore is incorrectly positive at this point;
2542 * the docs say it's undefined to reinit a semaphore that's already
2543 * init'd, so use timedwait to eat up any existing posts.
2545 while (qemu_sem_timedwait(&s->pause_sem, 1) == 0) {
2546 /* This block intentionally left blank */
2549 qemu_mutex_unlock_iothread();
2550 migrate_set_state(&s->state, *current_active_state,
2551 MIGRATION_STATUS_PRE_SWITCHOVER);
2552 qemu_sem_wait(&s->pause_sem);
2553 migrate_set_state(&s->state, MIGRATION_STATUS_PRE_SWITCHOVER,
2554 new_state);
2555 *current_active_state = new_state;
2556 qemu_mutex_lock_iothread();
2558 return s->state == new_state ? 0 : -EINVAL;
2562 * migration_completion: Used by migration_thread when there's not much left.
2563 * The caller 'breaks' the loop when this returns.
2565 * @s: Current migration state
2567 static void migration_completion(MigrationState *s)
2569 int ret;
2570 int current_active_state = s->state;
2572 if (s->state == MIGRATION_STATUS_ACTIVE) {
2573 qemu_mutex_lock_iothread();
2574 s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2575 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
2576 s->vm_was_running = runstate_is_running();
2577 ret = global_state_store();
2579 if (!ret) {
2580 bool inactivate = !migrate_colo_enabled();
2581 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
2582 if (ret >= 0) {
2583 ret = migration_maybe_pause(s, &current_active_state,
2584 MIGRATION_STATUS_DEVICE);
2586 if (ret >= 0) {
2587 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX);
2588 ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false,
2589 inactivate);
2591 if (inactivate && ret >= 0) {
2592 s->block_inactive = true;
2595 qemu_mutex_unlock_iothread();
2597 if (ret < 0) {
2598 goto fail;
2600 } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2601 trace_migration_completion_postcopy_end();
2603 qemu_savevm_state_complete_postcopy(s->to_dst_file);
2604 trace_migration_completion_postcopy_end_after_complete();
2608 * If rp was opened we must clean up the thread before
2609 * cleaning everything else up (since if there are no failures
2610 * it will wait for the destination to send it's status in
2611 * a SHUT command).
2613 if (s->rp_state.from_dst_file) {
2614 int rp_error;
2615 trace_migration_return_path_end_before();
2616 rp_error = await_return_path_close_on_source(s);
2617 trace_migration_return_path_end_after(rp_error);
2618 if (rp_error) {
2619 goto fail_invalidate;
2623 if (qemu_file_get_error(s->to_dst_file)) {
2624 trace_migration_completion_file_err();
2625 goto fail_invalidate;
2628 if (!migrate_colo_enabled()) {
2629 migrate_set_state(&s->state, current_active_state,
2630 MIGRATION_STATUS_COMPLETED);
2633 return;
2635 fail_invalidate:
2636 /* If not doing postcopy, vm_start() will be called: let's regain
2637 * control on images.
2639 if (s->state == MIGRATION_STATUS_ACTIVE ||
2640 s->state == MIGRATION_STATUS_DEVICE) {
2641 Error *local_err = NULL;
2643 qemu_mutex_lock_iothread();
2644 bdrv_invalidate_cache_all(&local_err);
2645 if (local_err) {
2646 error_report_err(local_err);
2647 } else {
2648 s->block_inactive = false;
2650 qemu_mutex_unlock_iothread();
2653 fail:
2654 migrate_set_state(&s->state, current_active_state,
2655 MIGRATION_STATUS_FAILED);
2658 bool migrate_colo_enabled(void)
2660 MigrationState *s = migrate_get_current();
2661 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO];
2664 typedef enum MigThrError {
2665 /* No error detected */
2666 MIG_THR_ERR_NONE = 0,
2667 /* Detected error, but resumed successfully */
2668 MIG_THR_ERR_RECOVERED = 1,
2669 /* Detected fatal error, need to exit */
2670 MIG_THR_ERR_FATAL = 2,
2671 } MigThrError;
2673 static int postcopy_resume_handshake(MigrationState *s)
2675 qemu_savevm_send_postcopy_resume(s->to_dst_file);
2677 while (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) {
2678 qemu_sem_wait(&s->rp_state.rp_sem);
2681 if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2682 return 0;
2685 return -1;
2688 /* Return zero if success, or <0 for error */
2689 static int postcopy_do_resume(MigrationState *s)
2691 int ret;
2694 * Call all the resume_prepare() hooks, so that modules can be
2695 * ready for the migration resume.
2697 ret = qemu_savevm_state_resume_prepare(s);
2698 if (ret) {
2699 error_report("%s: resume_prepare() failure detected: %d",
2700 __func__, ret);
2701 return ret;
2705 * Last handshake with destination on the resume (destination will
2706 * switch to postcopy-active afterwards)
2708 ret = postcopy_resume_handshake(s);
2709 if (ret) {
2710 error_report("%s: handshake failed: %d", __func__, ret);
2711 return ret;
2714 return 0;
2718 * We don't return until we are in a safe state to continue current
2719 * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or
2720 * MIG_THR_ERR_FATAL if unrecovery failure happened.
2722 static MigThrError postcopy_pause(MigrationState *s)
2724 assert(s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
2726 while (true) {
2727 QEMUFile *file;
2729 migrate_set_state(&s->state, s->state,
2730 MIGRATION_STATUS_POSTCOPY_PAUSED);
2732 /* Current channel is possibly broken. Release it. */
2733 assert(s->to_dst_file);
2734 qemu_mutex_lock(&s->qemu_file_lock);
2735 file = s->to_dst_file;
2736 s->to_dst_file = NULL;
2737 qemu_mutex_unlock(&s->qemu_file_lock);
2739 qemu_file_shutdown(file);
2740 qemu_fclose(file);
2742 error_report("Detected IO failure for postcopy. "
2743 "Migration paused.");
2746 * We wait until things fixed up. Then someone will setup the
2747 * status back for us.
2749 while (s->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
2750 qemu_sem_wait(&s->postcopy_pause_sem);
2753 if (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) {
2754 /* Woken up by a recover procedure. Give it a shot */
2757 * Firstly, let's wake up the return path now, with a new
2758 * return path channel.
2760 qemu_sem_post(&s->postcopy_pause_rp_sem);
2762 /* Do the resume logic */
2763 if (postcopy_do_resume(s) == 0) {
2764 /* Let's continue! */
2765 trace_postcopy_pause_continued();
2766 return MIG_THR_ERR_RECOVERED;
2767 } else {
2769 * Something wrong happened during the recovery, let's
2770 * pause again. Pause is always better than throwing
2771 * data away.
2773 continue;
2775 } else {
2776 /* This is not right... Time to quit. */
2777 return MIG_THR_ERR_FATAL;
2782 static MigThrError migration_detect_error(MigrationState *s)
2784 int ret;
2786 /* Try to detect any file errors */
2787 ret = qemu_file_get_error(s->to_dst_file);
2789 if (!ret) {
2790 /* Everything is fine */
2791 return MIG_THR_ERR_NONE;
2794 if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE && ret == -EIO) {
2796 * For postcopy, we allow the network to be down for a
2797 * while. After that, it can be continued by a
2798 * recovery phase.
2800 return postcopy_pause(s);
2801 } else {
2803 * For precopy (or postcopy with error outside IO), we fail
2804 * with no time.
2806 migrate_set_state(&s->state, s->state, MIGRATION_STATUS_FAILED);
2807 trace_migration_thread_file_err();
2809 /* Time to stop the migration, now. */
2810 return MIG_THR_ERR_FATAL;
2814 /* How many bytes have we transferred since the beggining of the migration */
2815 static uint64_t migration_total_bytes(MigrationState *s)
2817 return qemu_ftell(s->to_dst_file) + ram_counters.multifd_bytes;
2820 static void migration_calculate_complete(MigrationState *s)
2822 uint64_t bytes = migration_total_bytes(s);
2823 int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2824 int64_t transfer_time;
2826 s->total_time = end_time - s->start_time;
2827 if (!s->downtime) {
2829 * It's still not set, so we are precopy migration. For
2830 * postcopy, downtime is calculated during postcopy_start().
2832 s->downtime = end_time - s->downtime_start;
2835 transfer_time = s->total_time - s->setup_time;
2836 if (transfer_time) {
2837 s->mbps = ((double) bytes * 8.0) / transfer_time / 1000;
2841 static void migration_update_counters(MigrationState *s,
2842 int64_t current_time)
2844 uint64_t transferred, time_spent;
2845 uint64_t current_bytes; /* bytes transferred since the beginning */
2846 double bandwidth;
2848 if (current_time < s->iteration_start_time + BUFFER_DELAY) {
2849 return;
2852 current_bytes = migration_total_bytes(s);
2853 transferred = current_bytes - s->iteration_initial_bytes;
2854 time_spent = current_time - s->iteration_start_time;
2855 bandwidth = (double)transferred / time_spent;
2856 s->threshold_size = bandwidth * s->parameters.downtime_limit;
2858 s->mbps = (((double) transferred * 8.0) /
2859 ((double) time_spent / 1000.0)) / 1000.0 / 1000.0;
2862 * if we haven't sent anything, we don't want to
2863 * recalculate. 10000 is a small enough number for our purposes
2865 if (ram_counters.dirty_pages_rate && transferred > 10000) {
2866 s->expected_downtime = ram_counters.remaining / bandwidth;
2869 qemu_file_reset_rate_limit(s->to_dst_file);
2871 s->iteration_start_time = current_time;
2872 s->iteration_initial_bytes = current_bytes;
2874 trace_migrate_transferred(transferred, time_spent,
2875 bandwidth, s->threshold_size);
2878 /* Migration thread iteration status */
2879 typedef enum {
2880 MIG_ITERATE_RESUME, /* Resume current iteration */
2881 MIG_ITERATE_SKIP, /* Skip current iteration */
2882 MIG_ITERATE_BREAK, /* Break the loop */
2883 } MigIterateState;
2886 * Return true if continue to the next iteration directly, false
2887 * otherwise.
2889 static MigIterateState migration_iteration_run(MigrationState *s)
2891 uint64_t pending_size, pend_pre, pend_compat, pend_post;
2892 bool in_postcopy = s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE;
2894 qemu_savevm_state_pending(s->to_dst_file, s->threshold_size, &pend_pre,
2895 &pend_compat, &pend_post);
2896 pending_size = pend_pre + pend_compat + pend_post;
2898 trace_migrate_pending(pending_size, s->threshold_size,
2899 pend_pre, pend_compat, pend_post);
2901 if (pending_size && pending_size >= s->threshold_size) {
2902 /* Still a significant amount to transfer */
2903 if (migrate_postcopy() && !in_postcopy &&
2904 pend_pre <= s->threshold_size &&
2905 atomic_read(&s->start_postcopy)) {
2906 if (postcopy_start(s)) {
2907 error_report("%s: postcopy failed to start", __func__);
2909 return MIG_ITERATE_SKIP;
2911 /* Just another iteration step */
2912 qemu_savevm_state_iterate(s->to_dst_file,
2913 s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
2914 } else {
2915 trace_migration_thread_low_pending(pending_size);
2916 migration_completion(s);
2917 return MIG_ITERATE_BREAK;
2920 return MIG_ITERATE_RESUME;
2923 static void migration_iteration_finish(MigrationState *s)
2925 /* If we enabled cpu throttling for auto-converge, turn it off. */
2926 cpu_throttle_stop();
2928 qemu_mutex_lock_iothread();
2929 switch (s->state) {
2930 case MIGRATION_STATUS_COMPLETED:
2931 migration_calculate_complete(s);
2932 runstate_set(RUN_STATE_POSTMIGRATE);
2933 break;
2935 case MIGRATION_STATUS_ACTIVE:
2937 * We should really assert here, but since it's during
2938 * migration, let's try to reduce the usage of assertions.
2940 if (!migrate_colo_enabled()) {
2941 error_report("%s: critical error: calling COLO code without "
2942 "COLO enabled", __func__);
2944 migrate_start_colo_process(s);
2946 * Fixme: we will run VM in COLO no matter its old running state.
2947 * After exited COLO, we will keep running.
2949 s->vm_was_running = true;
2950 /* Fallthrough */
2951 case MIGRATION_STATUS_FAILED:
2952 case MIGRATION_STATUS_CANCELLED:
2953 case MIGRATION_STATUS_CANCELLING:
2954 if (s->vm_was_running) {
2955 vm_start();
2956 } else {
2957 if (runstate_check(RUN_STATE_FINISH_MIGRATE)) {
2958 runstate_set(RUN_STATE_POSTMIGRATE);
2961 break;
2963 default:
2964 /* Should not reach here, but if so, forgive the VM. */
2965 error_report("%s: Unknown ending state %d", __func__, s->state);
2966 break;
2968 qemu_bh_schedule(s->cleanup_bh);
2969 qemu_mutex_unlock_iothread();
2972 void migration_make_urgent_request(void)
2974 qemu_sem_post(&migrate_get_current()->rate_limit_sem);
2977 void migration_consume_urgent_request(void)
2979 qemu_sem_wait(&migrate_get_current()->rate_limit_sem);
2983 * Master migration thread on the source VM.
2984 * It drives the migration and pumps the data down the outgoing channel.
2986 static void *migration_thread(void *opaque)
2988 MigrationState *s = opaque;
2989 int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
2990 MigThrError thr_error;
2991 bool urgent = false;
2993 rcu_register_thread();
2995 s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2997 qemu_savevm_state_header(s->to_dst_file);
3000 * If we opened the return path, we need to make sure dst has it
3001 * opened as well.
3003 if (s->rp_state.from_dst_file) {
3004 /* Now tell the dest that it should open its end so it can reply */
3005 qemu_savevm_send_open_return_path(s->to_dst_file);
3007 /* And do a ping that will make stuff easier to debug */
3008 qemu_savevm_send_ping(s->to_dst_file, 1);
3011 if (migrate_postcopy()) {
3013 * Tell the destination that we *might* want to do postcopy later;
3014 * if the other end can't do postcopy it should fail now, nice and
3015 * early.
3017 qemu_savevm_send_postcopy_advise(s->to_dst_file);
3020 qemu_savevm_state_setup(s->to_dst_file);
3022 s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
3023 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
3024 MIGRATION_STATUS_ACTIVE);
3026 trace_migration_thread_setup_complete();
3028 while (s->state == MIGRATION_STATUS_ACTIVE ||
3029 s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
3030 int64_t current_time;
3032 if (urgent || !qemu_file_rate_limit(s->to_dst_file)) {
3033 MigIterateState iter_state = migration_iteration_run(s);
3034 if (iter_state == MIG_ITERATE_SKIP) {
3035 continue;
3036 } else if (iter_state == MIG_ITERATE_BREAK) {
3037 break;
3042 * Try to detect any kind of failures, and see whether we
3043 * should stop the migration now.
3045 thr_error = migration_detect_error(s);
3046 if (thr_error == MIG_THR_ERR_FATAL) {
3047 /* Stop migration */
3048 break;
3049 } else if (thr_error == MIG_THR_ERR_RECOVERED) {
3051 * Just recovered from a e.g. network failure, reset all
3052 * the local variables. This is important to avoid
3053 * breaking transferred_bytes and bandwidth calculation
3055 s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3056 s->iteration_initial_bytes = 0;
3059 current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3061 migration_update_counters(s, current_time);
3063 urgent = false;
3064 if (qemu_file_rate_limit(s->to_dst_file)) {
3065 /* Wait for a delay to do rate limiting OR
3066 * something urgent to post the semaphore.
3068 int ms = s->iteration_start_time + BUFFER_DELAY - current_time;
3069 trace_migration_thread_ratelimit_pre(ms);
3070 if (qemu_sem_timedwait(&s->rate_limit_sem, ms) == 0) {
3071 /* We were worken by one or more urgent things but
3072 * the timedwait will have consumed one of them.
3073 * The service routine for the urgent wake will dec
3074 * the semaphore itself for each item it consumes,
3075 * so add this one we just eat back.
3077 qemu_sem_post(&s->rate_limit_sem);
3078 urgent = true;
3080 trace_migration_thread_ratelimit_post(urgent);
3084 trace_migration_thread_after_loop();
3085 migration_iteration_finish(s);
3086 rcu_unregister_thread();
3087 return NULL;
3090 void migrate_fd_connect(MigrationState *s, Error *error_in)
3092 int64_t rate_limit;
3093 bool resume = s->state == MIGRATION_STATUS_POSTCOPY_PAUSED;
3095 s->expected_downtime = s->parameters.downtime_limit;
3096 s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s);
3097 if (error_in) {
3098 migrate_fd_error(s, error_in);
3099 migrate_fd_cleanup(s);
3100 return;
3103 if (resume) {
3104 /* This is a resumed migration */
3105 rate_limit = INT64_MAX;
3106 } else {
3107 /* This is a fresh new migration */
3108 rate_limit = s->parameters.max_bandwidth / XFER_LIMIT_RATIO;
3110 /* Notify before starting migration thread */
3111 notifier_list_notify(&migration_state_notifiers, s);
3114 qemu_file_set_rate_limit(s->to_dst_file, rate_limit);
3115 qemu_file_set_blocking(s->to_dst_file, true);
3118 * Open the return path. For postcopy, it is used exclusively. For
3119 * precopy, only if user specified "return-path" capability would
3120 * QEMU uses the return path.
3122 if (migrate_postcopy_ram() || migrate_use_return_path()) {
3123 if (open_return_path_on_source(s, !resume)) {
3124 error_report("Unable to open return-path for postcopy");
3125 migrate_set_state(&s->state, s->state, MIGRATION_STATUS_FAILED);
3126 migrate_fd_cleanup(s);
3127 return;
3131 if (resume) {
3132 /* Wakeup the main migration thread to do the recovery */
3133 migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_PAUSED,
3134 MIGRATION_STATUS_POSTCOPY_RECOVER);
3135 qemu_sem_post(&s->postcopy_pause_sem);
3136 return;
3139 if (multifd_save_setup() != 0) {
3140 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
3141 MIGRATION_STATUS_FAILED);
3142 migrate_fd_cleanup(s);
3143 return;
3145 qemu_thread_create(&s->thread, "live_migration", migration_thread, s,
3146 QEMU_THREAD_JOINABLE);
3147 s->migration_thread_running = true;
3150 void migration_global_dump(Monitor *mon)
3152 MigrationState *ms = migrate_get_current();
3154 monitor_printf(mon, "globals:\n");
3155 monitor_printf(mon, "store-global-state: %s\n",
3156 ms->store_global_state ? "on" : "off");
3157 monitor_printf(mon, "only-migratable: %s\n",
3158 ms->only_migratable ? "on" : "off");
3159 monitor_printf(mon, "send-configuration: %s\n",
3160 ms->send_configuration ? "on" : "off");
3161 monitor_printf(mon, "send-section-footer: %s\n",
3162 ms->send_section_footer ? "on" : "off");
3163 monitor_printf(mon, "decompress-error-check: %s\n",
3164 ms->decompress_error_check ? "on" : "off");
3167 #define DEFINE_PROP_MIG_CAP(name, x) \
3168 DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
3170 static Property migration_properties[] = {
3171 DEFINE_PROP_BOOL("store-global-state", MigrationState,
3172 store_global_state, true),
3173 DEFINE_PROP_BOOL("only-migratable", MigrationState, only_migratable, false),
3174 DEFINE_PROP_BOOL("send-configuration", MigrationState,
3175 send_configuration, true),
3176 DEFINE_PROP_BOOL("send-section-footer", MigrationState,
3177 send_section_footer, true),
3178 DEFINE_PROP_BOOL("decompress-error-check", MigrationState,
3179 decompress_error_check, true),
3181 /* Migration parameters */
3182 DEFINE_PROP_UINT8("x-compress-level", MigrationState,
3183 parameters.compress_level,
3184 DEFAULT_MIGRATE_COMPRESS_LEVEL),
3185 DEFINE_PROP_UINT8("x-compress-threads", MigrationState,
3186 parameters.compress_threads,
3187 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT),
3188 DEFINE_PROP_BOOL("x-compress-wait-thread", MigrationState,
3189 parameters.compress_wait_thread, true),
3190 DEFINE_PROP_UINT8("x-decompress-threads", MigrationState,
3191 parameters.decompress_threads,
3192 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT),
3193 DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState,
3194 parameters.cpu_throttle_initial,
3195 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL),
3196 DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState,
3197 parameters.cpu_throttle_increment,
3198 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT),
3199 DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState,
3200 parameters.max_bandwidth, MAX_THROTTLE),
3201 DEFINE_PROP_UINT64("x-downtime-limit", MigrationState,
3202 parameters.downtime_limit,
3203 DEFAULT_MIGRATE_SET_DOWNTIME),
3204 DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState,
3205 parameters.x_checkpoint_delay,
3206 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY),
3207 DEFINE_PROP_UINT8("x-multifd-channels", MigrationState,
3208 parameters.x_multifd_channels,
3209 DEFAULT_MIGRATE_MULTIFD_CHANNELS),
3210 DEFINE_PROP_UINT32("x-multifd-page-count", MigrationState,
3211 parameters.x_multifd_page_count,
3212 DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT),
3213 DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState,
3214 parameters.xbzrle_cache_size,
3215 DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE),
3216 DEFINE_PROP_SIZE("max-postcopy-bandwidth", MigrationState,
3217 parameters.max_postcopy_bandwidth,
3218 DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH),
3219 DEFINE_PROP_UINT8("max-cpu-throttle", MigrationState,
3220 parameters.max_cpu_throttle,
3221 DEFAULT_MIGRATE_MAX_CPU_THROTTLE),
3223 /* Migration capabilities */
3224 DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE),
3225 DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL),
3226 DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE),
3227 DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS),
3228 DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS),
3229 DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS),
3230 DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM),
3231 DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO),
3232 DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM),
3233 DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK),
3234 DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH),
3235 DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_X_MULTIFD),
3237 DEFINE_PROP_END_OF_LIST(),
3240 static void migration_class_init(ObjectClass *klass, void *data)
3242 DeviceClass *dc = DEVICE_CLASS(klass);
3244 dc->user_creatable = false;
3245 dc->props = migration_properties;
3248 static void migration_instance_finalize(Object *obj)
3250 MigrationState *ms = MIGRATION_OBJ(obj);
3251 MigrationParameters *params = &ms->parameters;
3253 qemu_mutex_destroy(&ms->error_mutex);
3254 qemu_mutex_destroy(&ms->qemu_file_lock);
3255 g_free(params->tls_hostname);
3256 g_free(params->tls_creds);
3257 qemu_sem_destroy(&ms->rate_limit_sem);
3258 qemu_sem_destroy(&ms->pause_sem);
3259 qemu_sem_destroy(&ms->postcopy_pause_sem);
3260 qemu_sem_destroy(&ms->postcopy_pause_rp_sem);
3261 qemu_sem_destroy(&ms->rp_state.rp_sem);
3262 error_free(ms->error);
3265 static void migration_instance_init(Object *obj)
3267 MigrationState *ms = MIGRATION_OBJ(obj);
3268 MigrationParameters *params = &ms->parameters;
3270 ms->state = MIGRATION_STATUS_NONE;
3271 ms->mbps = -1;
3272 qemu_sem_init(&ms->pause_sem, 0);
3273 qemu_mutex_init(&ms->error_mutex);
3275 params->tls_hostname = g_strdup("");
3276 params->tls_creds = g_strdup("");
3278 /* Set has_* up only for parameter checks */
3279 params->has_compress_level = true;
3280 params->has_compress_threads = true;
3281 params->has_decompress_threads = true;
3282 params->has_cpu_throttle_initial = true;
3283 params->has_cpu_throttle_increment = true;
3284 params->has_max_bandwidth = true;
3285 params->has_downtime_limit = true;
3286 params->has_x_checkpoint_delay = true;
3287 params->has_block_incremental = true;
3288 params->has_x_multifd_channels = true;
3289 params->has_x_multifd_page_count = true;
3290 params->has_xbzrle_cache_size = true;
3291 params->has_max_postcopy_bandwidth = true;
3292 params->has_max_cpu_throttle = true;
3294 qemu_sem_init(&ms->postcopy_pause_sem, 0);
3295 qemu_sem_init(&ms->postcopy_pause_rp_sem, 0);
3296 qemu_sem_init(&ms->rp_state.rp_sem, 0);
3297 qemu_sem_init(&ms->rate_limit_sem, 0);
3298 qemu_mutex_init(&ms->qemu_file_lock);
3302 * Return true if check pass, false otherwise. Error will be put
3303 * inside errp if provided.
3305 static bool migration_object_check(MigrationState *ms, Error **errp)
3307 MigrationCapabilityStatusList *head = NULL;
3308 /* Assuming all off */
3309 bool cap_list[MIGRATION_CAPABILITY__MAX] = { 0 }, ret;
3310 int i;
3312 if (!migrate_params_check(&ms->parameters, errp)) {
3313 return false;
3316 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
3317 if (ms->enabled_capabilities[i]) {
3318 head = migrate_cap_add(head, i, true);
3322 ret = migrate_caps_check(cap_list, head, errp);
3324 /* It works with head == NULL */
3325 qapi_free_MigrationCapabilityStatusList(head);
3327 return ret;
3330 static const TypeInfo migration_type = {
3331 .name = TYPE_MIGRATION,
3333 * NOTE: TYPE_MIGRATION is not really a device, as the object is
3334 * not created using qdev_create(), it is not attached to the qdev
3335 * device tree, and it is never realized.
3337 * TODO: Make this TYPE_OBJECT once QOM provides something like
3338 * TYPE_DEVICE's "-global" properties.
3340 .parent = TYPE_DEVICE,
3341 .class_init = migration_class_init,
3342 .class_size = sizeof(MigrationClass),
3343 .instance_size = sizeof(MigrationState),
3344 .instance_init = migration_instance_init,
3345 .instance_finalize = migration_instance_finalize,
3348 static void register_migration_types(void)
3350 type_register_static(&migration_type);
3353 type_init(register_migration_types);